Collaborative Transdisciplinary Educational Approaches in AI

Adriana Mihaela Coroiu

, Alina Delia C

alin

and Horea-Bogdan Mures¸an

Faculty of Mathematics and Computer Science, Babes¸-Bolyai University, Mihail Kog

alniceanu, Cluj-Napoca, Romania

Keywords:

AI, Machine Learning, Climate Change, Online Teaching, Transdisciplinary, Multidisciplinary.

Abstract:

In this paper we present a transdisciplinary approach towards teaching and applying AI methods, to mitigate

climate change related issues. The proposed method is a course with a student-centred approach, enabling

collaborative experiential learning and multi-disciplinary exploration with specialists from different ﬁelds of

expertise internationally. The study covered data collected through questionnaires, observations and evalua-

tion, and was proven to stimulate creativity, motivation and innovation in using AI to effectively solve real-life

problems, which is the aim of the course.

1 INTRODUCTION

One of the COVID-19 pandemic challenges and op-

portunities related to education was the requirement

to shift entirely to online teaching. This required new

and adapted methodologies to engage students in the

online environments however, this also posed new op-

portunities of connection and cooperation.

The educational system focuses on several dis-

ciplines which are studied in depth, but often the

bonding element between the subjects is missing or

is very weak. Therefore, educators and researchers

are still trying to ﬁgure out cross-domain solutions

to real-life problems. Recent tendencies are to-

wards new multi-domain or transdisciplinary and

inter-disciplinary studies, such as studying natural

sciences and the natural world as a whole from an

integral approach, rather than studying only separate

domains biology, geography, chemistry, physics and

social science. From this perspective, a study has

found that the lack of social science perspective on

the issues of climate change is lacking in science text-

books (Morris, 2014). This means the approach is

not stimulating the need to ﬁnd sustainable solutions

for the environment and green choices in the every-

day life. A successful example of multidisciplinar-

ity is neuroscience (Ruiter et al., 2012), looking at

the nervous system from all perspectives (physiology,

anatomy, molecular biology, developmental biology,

cytology, computer science).

https://orcid.org/0000-0001-5275-3432

https://orcid.org/0000-0001-7363-4934

https://orcid.org/0000-0003-4777-7821

Following this trend, the proposed computer sci-

ence course presented here is titled ”Artiﬁcial Intelli-

gence Models for Climate Change” and aims to bring

together multiple disciplines related to natural sci-

ences describing climate change issues and pollution

to inform recent advances in computer science (artiﬁ-

cial intelligence and machine learning, big data, sen-

sors, data storage) in the search of new solutions for

today’s real world problems. This also comes in line

with our University’s initiative ”to go GREEN”, en-

couraging active participation to reduce pollution and

ﬁnd ingenious solutions to Global Warming and Cli-

mate Change.

From a teaching methodology perspective, ap-

plied in the ﬁeld of computer science, recent stud-

ies have shown better educational outcomes in us-

ing collaborative learning and student-centred ap-

proaches (Lu et al., 2010) as opposed to teacher-

centred methodologies (Dias Canedo et al., 2017).

These involve providing materials that students go

through at their own pace, addressing more student

feedback and questionnaires in going deep into the

material (Grifﬁths et al., 2007) or using interactive

audio-visual content (Lu et al., 2010). However, al-

though these are not very new, the traditional teacher-

centred methodologies are still preserved, although

sometimes not as effective as desired.

Thus, this course was designed to use modern

teaching approaches which are more student-centred,

collaborative and critical thinking oriented, to in-

crease motivation and engagement, as well as the

overall learning experience of students.

260

Coroiu, A., C

alin, A. and Mure¸san, H.

Collaborative Transdisciplinary Educational Approaches in AI.

DOI: 10.5220/0011039500003182

In Proceedings of the 14th International Conference on Computer Supported Education (CSEDU 2022) - Volume 2, pages 260-267

ISBN: 978-989-758-562-3; ISSN: 2184-5026

2 ARTIFICIAL INTELLIGENCE

MODELS FOR CLIMATE

CHANGE - A NOVEL

TEACHING APPROACH

Besides the multi and transdisciplinary characteristic

of this course, combining multiple expertise on cli-

mate change and advanced artiﬁcial intelligence (AI)

techniques, its novelty is also in the modern educa-

tional approach towards collaborative and experien-

tial learning of students, teachers, and external ex-

perts in the ﬁeld. Moreover, the evaluation methodol-

ogy is also a reﬂective and collaborative one, involv-

ing also self-evaluation and inter-evaluation. While

many have considered and researched a multidisci-

plinary approach of the climate change issues from an

academic educational and research point of view, and

have proven that this is very much needed for good so-

lutions to arise, its success depends also on how this is

managed (Briguglio and Moncada, 2019), for which

reason the collaborative and experiential components

play very important roles.

2.1 About the Course

The course was held for the ﬁrst time in the aca-

demic year 2020-2021, second semester, addressing

third year Bachelor’s degree students who already had

studied the basics of Artiﬁcial Intelligence course.

There were 95 students enrolled from different spe-

cialisations: 54% Computer Science, 45% Mathemat-

ics and Computer Science, and 1% Mathematics), so

backgrounds were a bit different among the partici-

pants.

The general aim of the course was to identify so-

lutions for climate change using techniques and meth-

ods from Artiﬁcial Intelligence. The students would

enhance these abilities in the 12-weeks course:

• Identify current issues related to climate change

that could be addressed

• Model these identiﬁed problems

• Propose viable solutions in the form of software

applications that can solve at least partly the iden-

tiﬁed problems

The best keywords to describe this course are:

innovation, applicability, collaboration, development

and transdisciplinarity.

2.1.1 Innovation and Applicability

The subjects addressed were still recent and full of

challenges, both from the perspective of data sources

involved in the analysed subject (Earth, Environment,

Agriculture, Society) and from the point of view of

the Artiﬁcial Intelligence Methods that can be ap-

plied. The implemented solutions were required to

be robust, accurate with minimal compromise, and to

have a positive impact on the health of the people and

of the environment. Computer Science and Artiﬁcial

Intelligence, through the mathematical device behind

them and the existent tools, are able to provide useful

and concrete solutions to real problems of society and

humanity.

2.1.2 Collaboration and Development

The course was student-centred oriented and dy-

namic. We started from the state-of-the-art of re-

search literature for each domain, then involving a

specialist (invited to get more insight in the analysed

issues) that presented a very concrete problem. Spe-

cialists and students have discussed the requirements

and proposed solutions in the form of software appli-

cations that were developed during the labs. The labo-

ratory involved group collaborative work between 4-6

students.

2.1.3 Transdisciplinar

As the name suggests, the course involved several do-

mains. This component is also supported by the Uni-

versity resources, which allow us to invite specialists

from other faculties or departments, even industry ex-

perts or NGOs activists internationally, to share their

knowledge and expertise with students.

2.2 Platforms and Tools Utilised

The course was held in the context of online teaching

imposed by the COVID19 pandemic, using the on-

line platforms Moodle (for content sharing in terms

of materials and theory, and evaluation), Microsoft

Teams University platform or the Zoom platform (for

open discussions, especially involving guest speakers,

presentations sharing and feedback) and JupyterLab

(an interactive development environment for develop-

ing the projects, using Python as programming lan-

guage, machine learning, related packages for data

pre-processing, model training and testing) (Perez

and Granger, 2015). As an online development tool,

including libraries (Tensorﬂow, Keras, ScikitLearn),

it is well suited for collaborative programming and

group projects, facilitating code sharing, but also

models and obtained results.

Collaborative Transdisciplinary Educational Approaches in AI

261

2.3 Teaching Methodologies

As teaching methodology, the course is taking a shift

towards student-centred approach through coopera-

tive learning, inquiry based learning and experien-

tial learning. Collaborative learning is performed

from three perspectives: perspective teacher-student,

perspective student-student and perspective student-

specialist (specialist meaning a guest invited to share

knowledge and expertise in a different ﬁeld). The

teacher’s role starts with facilitating the introduction

of new concepts of AI methodologies, but it mostly

continues as a facilitator and delegator, involving crit-

ical thinking and discussions between students and

external experts in various transdisciplinary ﬁelds re-

lated to climate change.

This is based on the approach toward innovation

and cutting-edge research, aiming to stimulate cre-

ativity, interest, critical thinking and involvement in

actual solving of real life problems that are close to

the experience of a student. Reaching into the intrin-

sic motivation of students, will engage a much deeper

learning approach (Baeten et al., 2010) and increased

results, especially if they satisﬁed with the course

overall. This can be achieved by selecting content that

is of interest to them (which was performed in this

case at the beginning of the course by using a ques-

tionnaire). The votes are distributes as such: 29.9%

AI for Society, 25.3% AI for Agriculture, 23.7% AI

for Earth, and 21.1% AI for Environment.

From this point of view, experiential learning is

the basis not only from the learning during the course

perspective, but also from an evaluation perspective,

meaning students are asked to reﬂect on their work

and their colleagues work and evaluate the innovation

and performance of the solutions. Therefore, evalu-

ation is also collaborative in this sense, between stu-

dents, teacher and specialists in other ﬁelds.

Table 1: Research questions.

Questions for analysing a research paper:

1. What is the main problem addressed?

2. What was done before, and how does this paper

improve on it?

3. What is the one cool technique/idea/ﬁnding

that was learned from this paper?

4. What part was difﬁcult to understand?

5. What generalisation or extension of the paper

could be done?

The laboratory work is based on a kinestethic

learning (involving hands-on experience). Group

projects involve differentiated instruction based on

speciﬁc aims of the chosen project and requirement,

as well as based on the team capabilities, starting with

literature review, debating possible improvements and

how relevant are the results (the questions are pre-

sented in Table 1). It might involve also a level of

expeditionary leaning (engaging with experts to learn

even more on the subject, understand the problem and

decide the action).

After identifying a relevant research paper and

performing a critical analysis, the next challenge is

ﬁnding relevant data (or collecting/generating it). We

tried focusing on data collected in Romania or Eu-

rope, as research clearly shows that educational in-

terventions are most successful when they focus on

local, tangible, and actionable aspects of sustainable

solutions (Anderson, 2012). This way, the aim was to

engage students in a local tangible problem that they

can solve with AI, emphasising their crucial role in

it, but also to educate them to remember the issues of

the environment beyond the computer science course,

and keep a focused attitude towards protecting the en-

vironment within all aspects of their life. The ﬁnal

datasets used in the projects are from eleven different

sources, as presented in Figure 1.

Figure 1: Datasets sources.

Next, the general steps required when apply-

ing artiﬁcial intelligence algorithms (as presented in

detail in Section 2.5) were given as guidance for

project development. This guidance was useful for

the groups to work together and collaborate, espe-

cially as groups were mixed (more computer science

oriented or more mathematical oriented background)

to encourage communication and sharing of differ-

ent perspectives, involving an important component

of knowledge transfer between the team members of

each group.

Group projects which are problem-based learning

pose several advantages: the development of criti-

cal thinking and creative skills, the improvement of

CSEDU 2022 - 14th International Conference on Computer Supported Education

262

problem solving abilities, increased student motiva-

tion, better knowledge sharing in challenging situa-

tions and it is a form of experiential learning.

2.4 Subjects Addressed by the Course

Half of the course included technical content re-

lated to optimising AI methods (for example Features

Selection/Extraction Techniques, Ensemble Learning

Methods or Hyperparameter Techniques), and the

other half involved a guest speaker in one of the inter-

est domains voted by students about climate change,

which are presented below.

2.4.1 Light Pollution

This presentation was performed by Mihai Cuibus,

Software Engineer, member of the Romanian Society

for Cultural Astronomy, and lead on the Light Pol-

lution department. He brought to light a new topic

with high impact which is still rarely discussed in our

country, by presenting the impact of light pollution

at a national/local level, as well as globally, for the

human health, as well as for the natural world and

its biodiversity. After discussing the different issues

involved and possible solutions from a Computer Sci-

ence point of view, some speciﬁc tasks were identiﬁed

mostly related to image processing and analysis:

• Detecting light sources in an image

• Spectral analysis of an image (colour tempera-

ture)

• Predicting the extension of light pollution based

on satellite images (such as Figure 2) and corre-

lation with health issues related to light pollution

for those regions; also correlation with air pollu-

tion on those areas

• Mapping the streets of a neighbourhood with a

luxmeter or SQL or spectrometer and correlating

these with existing maps

• Design of intelligent street lights that can be re-

motely controlled, function based on time of day,

trafﬁc intensity, or geographic location, adapting

to the user behaviour to reduce light where it is

not used, and automatic adjustment of the light

intensity based on certain conditions

2.4.2 Sustainability and Society

This presentation was held by Sorina Avadanei, mas-

ter student at the Anthropology Department from

Durham University. She presented the complex prob-

lem of sustainability as a result of technological ad-

vancement and the people that intervene in the en-

vironment. She compared how people relate to this

Figure 2: Light pollution sky map for Cluj-Napoca and sur-

roundings generated in May 2021 using the new atlas of

artiﬁcial of night sky brightness (Microsoft, 2021; Falchi

et al., 2016). Warmer colours indicate a higher degree of

light pollution.

in Romania or UK and presented some relevant ideas

that can be tackled by Computer Science, such as de-

veloping an application which based on a map dis-

plays the closest recycling units to the user.

2.4.3 Agent Green

Agent Green is an NGO for the protection of the en-

vironment created in 2009 in Romania. The organisa-

tion investigated environment crimes and tries to ex-

pose them strategically, promoting solutions for bio-

diversity conservation and the assurance of a health

environment for the future generations. Veronica Tul-

pan and Raluca Nicolae have been involved in the pre-

sentation, discussing their national projects, the actual

stage of forests in Romania and the impact of uncon-

trolled deforestation on climate change. Other issues

discussed were the increasing consumption of meat

and the high impact of methane emissions from in-

tensive breeding. The proposed ideas were

• Identifying deforestation areas and levels by using

drone/satellite images

• Identifying the age of a tree based on the number

of extracted samples

• Creating an app to encourage nature lovers to

meet for planting trees, or perhaps extinguishing

ﬁres or exploring

2.4.4 Human Ecology and the Impact of Climate

Change on Biodiversity

This presentation performed by Alexandru Stermin,

Lecturer at the Faculty of Biology, highlighted the

faculty’s projects to protect the environment and iden-

tify methods that are able to automatise this process.

Collaborative Transdisciplinary Educational Approaches in AI

263

The solutions envisioned are:

• Identifying the area of separation between taiga /

steppe based on the current images and compar-

ing them with those existing for several decades,

to identify changes in their structure (their retreat

faster and faster to the north)

• Automatic identiﬁcation of birds and framing in a

certain species based on a picture / video taken

• Automatic analysis of information related to the

total number of birds of a certain species (see Fig-

ure 3) and plants in a given area and prediction for

the coming years (it was found that there are ar-

eas where bird populations and certain plants are

constantly declining - also based on the harmful

effects of humans on the environment)

Figure 3: Identifying and counting birds from images.

• Creating a mobile app that warns us how much

harm we do to the environment through daily ac-

tivity, for example: recording the sound of run-

ning water used for the duration of brushing teeth,

or identifying the engine noise of a car for travel

and providing notiﬁcations in real-time or at the

end of the day related to the risk that our day rep-

resents for the environment.

2.4.5 Microsoft Initiatives for the Environment

This presentation performed by Rusen Daniel and Lu-

cian Ungureanu was based on the Microsoft AI for

Earth projects. They also mentioned the reasons for

being a sustainable company, such as 100% recycling

policy or garden roofs, and offered free credits access

to students for Microsoft Azure.

2.5 Steps Required When Applying

Artiﬁcial Intelligence Algorithms

1. Understanding the Domain: In order to make

an informed decision on what AI algorithm is best

for a given task, understanding the currently avail-

able data that can serve as input is mandatory. As

an AI model processes data, it produces an output

which can take several forms (array of probabili-

ties, number, feature map etc.). Choosing an AI

algorithm should take into consideration the pro-

vided output.

2. Data Gathering: Part of developing a good AI

model is selecting appropriate data to train it on.

The data should include samples for each category

and, ideally, a balanced number of samples per

category. In some cases, such datasets already ex-

ist, however, in all other cases the dataset must be

built. To improve the quality of a dataset, a couple

of techniques can be applied: data sensitisation

and preprocessing (Famili et al., 1997; Perez and

Wang, 2017). Data sensitisation involves identi-

fying and eliminating data points that are not rele-

vant to the studied problem or that contain invalid

attributes. Preprocessing data refers to a num-

ber of algorithms that can enlarge the dataset by

adding new data points generated based on the ex-

isting ones (eg. for image datasets, image ﬂips and

rotations can achieve this goal).

3. Exploratory Data Analysis: Regardless of how

the dataset is obtained, overview information such

as total count, count per class, minimum, maxi-

mum, average and standard deviation (where ap-

plicable) can help further understanding of dataset

characteristics (Famili et al., 1997). The identiﬁ-

cation of existent correlations within the dataset

can also help in simplifying the training process

by reducing the amount of processed features.

4. Determining the AI Algorithm: Choosing the

right algorithm to train a model requires careful

consideration of available input data characteris-

tics as well as target result type(Dey, 2016). If

the desired outcome is to identify patterns in un-

labelled data and grouping data points based on

their feature, the go to method will be an un-

supervised learning algorithm. Among the most

widespread unsupervised learning algorithm is

clustering, which attempts to partition a set of data

points into clusters by minimising the distance be-

tween data points of the same cluster and by max-

imising the distance between data points of dif-

ferent clusters. Should labelled data be available,

then supervised learning can be employed. This

CSEDU 2022 - 14th International Conference on Computer Supported Education

264

class of machine learning algorithms can predict

either a continuous value (regression) or can iden-

tify to which class/classes a data point belongs to

(classiﬁcation).

5. Training the Model: Once the dataset is ob-

tained or created and the AI method is chosen, the

training process can begin. It is recommended to

split the dataset into three subsets: train - valida-

tion - test (Xu and Goodacre, 2018). The train

subset will be used by the AI algorithm to adjust

the model’s weight such that the values of perfor-

mance metrics (accuracy, loss, F1-score, etc) tend

towards the optimum. The validation subset will

be used at the end of a training epoch (ie. when

the training algorithm has iterated over the entire

train subset) to evaluate the current performance

of the model. Based on this, hyper-parameters

such as learning rate can be adapted during train-

ing, while a stagnation in performance can lead to

an early stop of the process. The validation sub-

set must be distinct from the train subset as it can

easily indicate if the model is unable to generalise

on new data, issue known as over-ﬁtting.

6. Evaluating the Model: The test subset is not

involved in training/validating the model and typ-

ically contains samples that are meant to repre-

sent ”real-world” data points. As such, this subset

is the most appropriate for evaluating the perfor-

mance of the trained model and it serves as an in-

dicator to how well can the model be used in a

practical application.

7. Validating, Visualising and Interpreting the

Results: To gain insight on the performance of

a model, the evaluation results can be aggregated

into tables and plotted on charts. This is a more

”human readable” way to analyse data and makes

it more simple to interpret it. The evolution over

time of the performance of the model during train-

ing can help indicate if the training is too long

(the model saturated early and no longer learns)

or is too short (the model keeps improving every

epoch). In the case of classiﬁers, a confusion ma-

trix can quickly show if the model can distinguish

properly between the classes, and if not, which are

the classes that are considered very similar.

8. Synthesising the Results: Once results are or-

ganised, they can be summarised to form a con-

clusion. This should contain observations regard-

ing the problem that the model solves side by side

with other results from the literature. The key dif-

ferences between the proposed method and the ex-

isting methods should be highlighted, with both

advantages and disadvantages. Finally, further

improvement directions can be suggested based

on observed weaknesses of the model and new di-

rections for expanding the model can be provided.

2.6 Course Evaluation Methods and

Results

The ﬁnal evaluation for the course involved:

• 30% Self-evaluation of the project team

• 30% Evaluation of the ﬁnal project (involving a

presentation and a demo or an application and a

descriptive report)

• 30% Evaluation of the work by other teams

• 10% Activity during the semester at the lab

Each team had to evaluate their strategy of work-

ing, roles distribution, internal task assignment and

development of solutions, and to propose a grades

based on these aspects. The ﬁnal project was based

on an application and a descriptive report (describing

the motivation, the problem to be solved, theoretical

elements used, similar approaches in literature, pre-

sentation of the methodology chosen for the solution

and the application), which was aimed at:

• solving a problem related to the climate change,

environment, or pollution using one of the ad-

vanced AI methods presented;

• use the recommended steps in the solution devel-

opment;

• be innovative and refer to the state-of-the-art pub-

lished in the last 5 years and no later then this.

It was mandatory that each team will participate

at the presentations of other teams, to reﬂect on other

approaches, to ask questions and then to evaluate the

team projects. Based on the average of the results for

the team, the students should allocate to each of them

(inside the team) a particular grade, thus obtaining the

average which is the team’s evaluation grade. In this

way, they have the opportunity to correctly reﬂect the

individual grade according to the implication within

the project. They demonstrated a fair analysis, the re-

sults recompensing those students who worked harder

than others.

The projects and solutions proposed by students

are related to: drought and ﬂooding prediction,air

quality classiﬁcation and prediction, fruit recognition

from images, detecting plants and leaf diseases, an-

imal monitoring by image recognition, storm detec-

tion, deforestation detection, forest ﬁre prediction,

intelligent plant irrigation, greenhouse monitoring,

weather and temperature prediction, city metabolism

analysis, care homes and orphanages classiﬁcation

Collaborative Transdisciplinary Educational Approaches in AI

265

based on needs and priorities, trafﬁc prediction in Eu-

rope’s big cities, and household electricity consump-

tion.

The intelligent algorithms, models, methods and

datasets involved are: LSTM, CNN, K-Means, Clus-

ters, PCA, ANN, Regression, SVM, Random Forest,

DeepWeeds, Inception-v3, ResNet-50, and the data-

sources presented previously in Figure 1.

Figure 4: Distribution of grades for each component of the

evaluation.

The grades obtained by each team are presented in

Figure 4, including each component of the evaluation.

Comparing the three types of team project evaluation

(self-evaluation, inter-evaluation and teacher evalua-

tion) as seen in Figure 5, we can observe that inter-

evaluation and teacher evaluation are similar and have

less variance, as opposed to self-evaluation, which has

a high variance, and surprisingly, with a lower mean

(9.02) then inter-evaluation mean (9.46) and teacher

evaluation mean (9.12). This reﬂects the different ex-

pectation that each team had related to the project’s

success.

3 EVALUATION AND BENEFITS

OF THE APPROACH

At the end of the semester, it seems that the main ad-

vantage of the course methodology was represented

by the mixture of technical and non-technical presen-

tations. On one hand, the presentations delivered by

the external guests in different domains about envi-

ronment issues were eye-opening for students. The

process of being aware of unseen problems of earth

and the environment brings a new perspective for

most of them (they even mention ”we didn’t know

about this issue ...”). On the other hand, the technical

Figure 5: Self evaluation, inter-team evaluation and teacher

evaluation for each project team.

presentations about literature and AI techniques clar-

iﬁed the main and useful ML approaches for ﬁnding

solutions for such issues, offering a set of rules for an

AI-based project.

At the end of the activity, the students were asked

to provide antonymous feedback using Google forms.

Most of them mentioned they learn a lot of new things

from our guests. They express their gratitude for the

opportunity and they mention that this discipline was

one of the most interesting from this semester and

they performed an awesome choice to enrol with this

one. Regarding the evaluation process, they provided

feedback that the idea of self-evaluation and the idea

of evaluating other teams allow them to feel that their

opinion is important and valuable. They also said that

they were more conﬁdent in expressing interest in pre-

sentations and addressing questions.

In what concerns the project development, they

appreciated they have worked for solutions than can

be extended and used in real-life, and this was not

only a school project. Some also mentioned that the

discussions with the guests was introducing them to

the industry context of working with the clients into a

company, in order to ﬁgure out the requirements from

a non-technical person and translate it into a computer

science solution.

In terms of drawbacks, students highlighted that

their main problem was that the course was held in

the last semester of bachelor’s degree, by which time

they have already chosen their degree title, thus hav-

ing limited time to invest in this course.

The students also emphasised that this was the

third discipline in three years in which they were re-

quired to work in a team. In this sense, the idea of

collaboration while also keeping in mind the individ-

ual task and the collective task was a real challenge

and they were proud they succeed in it.

CSEDU 2022 - 14th International Conference on Computer Supported Education

266

4 CONCLUSIONS

The presented approach of the course ”Artiﬁcial Intel-

ligence for Climate Change” was new into the faculty

and for the students, but the feedback and the interest

from both students and specialists invited show that

the idea can be of success also in the next years.

To the best of our knowledge, this approach is

unique among the universities of Romania, in par-

ticular, computer science faculties. Globally how-

ever, this program is implemented across several uni-

versities such as: AI for Social Good from Stan-

ford, Climate Change and AI from Oxford and AI for

the study of Environmental Risks from University of

Cambridge.

In the scientiﬁc community there is a great interest

for this topic, ”AI for Good” being the largest confer-

ence to date (https://aiforgood.itu.int/). There are also

working groups consisting of researchers and special-

ists to support the ﬁght against climate change (Fo-

cus Group on Environmental Efﬁciency for Artiﬁcial

Intelligence and other Emerging Technologies, Focus

Group on AI for Natural Disaster Management, Focus

Group on AI for Health, etc.).

As future work directions, our intention is to ex-

tend the list of specialists from different areas, in-

crease the numbers of AI teachers and to organise

some workshop during the laboratory work activity -

engaging also experts from other ﬁelds to offer feed-

back in real-time for a new implemented feature.

ACKNOWLEDGEMENTS

The work presented became real with the help of sup-

portive colleagues, extremely passionate guests, and

highly involved students. Therefore we would like to

express our gratitude to Professor Motogna Simona,

Lecturer Alexandru Stermin and specialists Mihai

Cuibus (Romanian Society for Cultural Astronomy),

Veronica Tulpan and Raluca Nicolae (Agent Green),

Sorina Avadanei (Durham University), Daniel Rusen

and Lucian Ungureanu (Microsoft Romania).

The publication of this paper was supported by the

2021 Development Fund of the Babes¸-Bolyai Univer-

sity (UBB).

REFERENCES

Anderson, A. (2012). Climate change education for mitiga-

tion and adaptation. Journal of Education for Sustain-

able Development, 6(2):191–206.

Baeten, M., Kyndt, E., Struyven, K., and Dochy, F. (2010).

Using student-centred learning environments to stim-

ulate deep approaches to learning: Factors encourag-

ing or discouraging their effectiveness. Educational

Research Review, 5(3):243–260.

Briguglio, L. and Moncada, S. (2019). The beneﬁts and

downsides of multidisciplinary education relating to

climate change. Climate Change and the Role of Ed-

ucation, pages 169–187.

Dey, A. (2016). Machine learning algorithms: a review.

International Journal of Computer Science and Infor-

mation Technologies, 7(3):1174–1179.

Dias Canedo, E., Santos, G. A., and Andrade de Fre-

itas, S. A. (2017). Analysis of the teaching-learning

methodology adopted in the introduction to computer

science classes. In 2017 IEEE Frontiers in Education

Conference (FIE), pages 1–8.

Falchi, F., Cinzano, P., Duriscoe, D., Kyba, C. C., Elvidge,

C. D., Baugh, K., Portnov, B. A., Rybnikova, N. A.,

and Furgoni, R. (2016). The new world atlas of

artiﬁcial night sky brightness. Science advances,

2(6):e1600377.

Famili, A., Shen, W.-M., Weber, R., and Simoudis, E.

(1997). Data preprocessing and intelligent data anal-

ysis. Intelligent data analysis, 1(1):3–23.

Grifﬁths, G., Oates, B. J., and Lockyer, M. (2007). Evolving

a facilitation process towards student centred learning:

A case study in computing. Journal of Information

Systems Education, 18(4):459.

Lu, Z., Hou, L., and Huang, X. (2010). A research on a

student-centred teaching model in an ict-based english

audio-video speaking class. International Journal

of Education and Development Using ICT, 6(3):101–

123.

Microsoft (2021). Light pollution maps.

Morris, H. (2014). Socioscientiﬁc issues and multidisci-

plinarity in school science textbooks. International

Journal of Science Education, 36(7):1137–1158.

Perez, F. and Granger, B. E. (2015). Project jupyter: Com-

putational narratives as the engine of collaborative

data science. Retrieved September, 11(207):108.

Perez, L. and Wang, J. (2017). The effectiveness of data

augmentation in image classiﬁcation using deep learn-

ing. arXiv preprint arXiv:1712.04621.

Ruiter, D. J., van Kesteren, M. T., and Fernandez, G. (2012).

How to achieve synergy between medical education

and cognitive neuroscience? an exercise on prior

knowledge in understanding. Advances in health sci-

ences education, 17(2):225–240.

Xu, Y. and Goodacre, R. (2018). On splitting training

and validation set: A comparative study of cross-

validation, bootstrap and systematic sampling for es-

timating the generalization performance of supervised

learning. Journal of Analysis and Testing, 2.

Collaborative Transdisciplinary Educational Approaches in AI

267